Instant field aligned meshes

Very nice, it has great potential. And all this in just 2.5MB.
Thanks for sharing and thanks to the developers!

Some basic tests without looking much the tutorial, I am very happy :slight_smile:

Instant meshes crash wen I minimized the windows were i can report bugs “win 7 64”

It seams the bleu one had a more constant “tile” mesh size then the orange one.
Wouldnt a lower mesh count be better in these kind of problems ea never mind there are some tris
Also for the bleu one, could you select all meshes and their face type to smooth
It seams zbrush had it done too.

Also whats the effect of a laplacian smooth modifier on these models i’m curious

  • The zremesh is more accurate and keep the topology flow curvature and much more cleaner mesh
  • the instant mesh don’t take into account the curvature and flow of the mesh but it have a potential

Here’s the blend file which also includes the original model and the ZRemesher model with roughly the same number of faces as the Instant Mesh model

Wouldnt a lower mesh count be better in these kind of problems ea never mind there are some tris

Here’s the same with each mesh at about 5K polys


remesh_test_5K.blend (597 KB)

Zremesh gives better results.

Although we have to take into account how much the mastering of brush tools can improve the result in Instant Meshes.
(so far, I have not been able to use them)


The first commit is 2 weeks old and the programmer behind it is Wenzel who did the mitsuba renderer. So I’m sure it will mature pretty well if you give him some nice feedback and some time :slight_smile:

Looking at it, the IFAM algorithm is pretty good… but have to say, based on the examples provided that it still falls well short of ZRemesher. It would be hard to argue that the topology in the examples provided by Mr Marklew show ZBrush is doing way better.

That’s not a total loss though because IFAM is still pretty good, though not yet the replacement for manual retopology that ZRemesher can be, and it may be possible to for it to give better results with further tweaking of the algorithm and/or parameters it uses. I don’t know enough about the algorithm to speculate further but it’s sparked enough interest to at least grab the paper & code for perusing later. :slight_smile:

This is looking nice, but I wish the algorithm could create a pure quad mesh even if the mesh is not that evenly spaced… But as a remesher for dytopo meshes this could work very good, but not for automatic retopology.

the Zbrush version is better, but at least the instant version is (to my eye) competitive-ish. Good tool to know about, esp. for lazy people like me who try to avoid retopo at all costs anyway…

Eh, if you need perfect topology, then there’s no substitute for manual labor anyway, ZRemesher only gives you what you really want with a lot of luck and a great deal of tweaking.

For every other purpose, this is good enough. Like I said earlier, the only thing really missing here is density control.

Yes the results are good even if it’s not as good as zremesher.
It’s in a good way and it will be great if it will be integrated to blender.

I don’t know where people get the idea that Zremesher output is better. The only problems here on the other mesh is for sculpting in multi-res (Triangles). Richard pointed out the zremesher used default settings that means “adaptive” is enabled . For best comparison we should disable adaptive setting.

As for zremesher being better, as you can see in the picture below the red circles show the problematic areas. Zremesher has non uniform polygon sizes on the generated mesh and has different densities on different parts of the mesh.

This produces resolution mismatch when the model is subdivided , one part of the model can hold much more detail while the other part can not - this is also true for vertex painting since the subdivided sculpt wont have uniform number of polys all around.

Also uniform polygon distribution allows greater quality when transposing the model (if need be).

Main reason why zremeshes has this kind of output is because it tries to follow edge flow (Adaptive - Freeze Border - Curve Guides - Paint Guides) and keep the target poly count. So for a better comparison adaptive and similar helpers need to be turned off. Alternatively you can use all the helpers of field aligned remesher and use all the helpers of zremesher.

One last advantage of field aligned remesher that I noticed is that helpers allow interactive feedback. You can see an approximation of the distribution. In zremesher you need to play with parameters (guesswork) and wait for the complete result.

I couldn’t be happier, and it’s working great here. The orientation comb is easy and works really well (would be nice as a hotkey though, right now just turn it on, draw, then turn it off). Honestly, this does more than I need it to. Definitely, the ability to define lower-res (non-uniform) density for areas of the mesh would be the ultimate, but even without that, this is just really useful and I can make due as is.

Who-ever made this, if you see this, thank you so much!

Just FYI, (and it’s not nearly as easy to use) there is another open source tool called MeshLab that might be use-ful to getting rid of unnecessarily dense areas in a semi-automatable way, especially for triangle based in-game meshes if you can find the right settings/tools and a good default/go to set (there are a lot of options).

Well, zremesher sacrifices polygon uniformity for fewer singularities. It’s a trade off.

I wouldn’t call the triangles a problem, just subdivide the mesh once and they’re gone (but you still get a trivalent singularity there).

What exactly is a singluarity?

Well, it’s a point where the surface, or field, becomes degenerate. Finding singularities is part of most of these nice retopology algorithms. I don’t really want to get into math here, so to keep it simple, it’s a vertex with less or more than 4 edges attached, also known as a pole.

Can’t test as i don’t have 64bits OS, but it sure looks great.

In the comparison screenshot, it does not look as good as the zbrush remesher, but it looks already a very lot better than the blender remesher as i don’t see the “lose all details unless you skyrocket the remeshing polycount and put losange patterns everywhere” thing that the blender remesher does.

I hope this one will be able to be implemented into Blender one day.